what childhood characteristics predict psychological resilience to economic shocks in adulthood?

Journal of Economic Psychology 45 (2014) 84–101

Contents lists available at ScienceDirect

Journal of Economic Psychology

journal homepage: www.elsevier .com/ locate / joep

What childhood characteristics predict psychological resilienceto economic shocks in adulthood?

http://dx.doi.org/10.1016/j.joep.2014.08.0030167-4870/� 2014 Elsevier B.V. All rights reserved.

⇑ Address: Centre for Economic Performance, London School of Economics, Houghton Street, London WC2A 2AE, UK. Tel.: +44 (0)7990 815924E-mail address: [email protected]

Nattavudh Powdthavee ⇑Centre for Economic Performance, London School of Economics, UKMelbourne Institute of Applied Economic and Social Research, University of Melbourne, Australia

a r t i c l e i n f o

Article history:Received 2 January 2014Received in revised form 12 June 2014Accepted 24 August 2014Available online 2 September 2014

JEL classification:D03I19J64

PsycINFO classification:2900

Keywords:ResilienceHappinessUnemploymentChildhoodBHPS

a b s t r a c t

This paper investigates whether people’s psychological resilience to one of the most impor-tant economic shocks – job loss – can be predicted using early childhood characteristics.Using a longitudinal data that tracked almost 3000 children into adulthood, we showed thatthe negative effect of unemployment on mental well-being and life satisfaction is signifi-cantly larger for workers who, as adolescents, had a relatively poor father-child relationship.Maternal unemployment, on the other hand, is a good predictor of how individuals react psy-chologically to future unemployment. Although the results should be viewed as illustrativeand more research is needed, the current article provides new longitudinal evidence that psy-chological resilience to job loss may be determined early on in the life cycle.

� 2014 Elsevier B.V. All rights reserved.

1. Introduction

There is a growing body of literature in economics over the last decade on the causes and consequences of human psy-chological resilience, i.e. the evidence of a lower volatility in well-being and a quick return to a baseline level of happinessfollowing a significantly bad life event (Graham & Oswald, 2010; Perez-Truglia, 2012; Rayo & Becker, 2007). This recent surgeof interest among economists is fuelled by the releases of new longitudinal evidence of people adapting quickly and com-pletely in terms of mental well-being and life satisfaction to negative life shocks, including adaptation to unemployment,disability, and bereavement (Clark, Diener, Georgellis, & Lucas, 2008; Frijters, Johnston, & Shields, 2011; Oswald &Powdthavee, 2008), as well as by the potential implications of how people adapt to bad shocks in life have in public policyand welfare evaluation (Frederick & Loewenstein, 1999; Layard, 2006; Loewenstein & Ubel, 2008).

According to psychologist George Bonanno (2004), psychological resilience in adulthood is defined as,

.

http://crossmark.crossref.org/dialog/?doi=10.1016/j.joep.2014.08.003&domain=pdf

http://dx.doi.org/10.1016/j.joep.2014.08.003

mailto:[email protected]

http://dx.doi.org/10.1016/j.joep.2014.08.003

http://www.sciencedirect.com/science/journal/01674870

http://www.elsevier.com/locate/joep

N. Powdthavee / Journal of Economic Psychology 45 (2014) 84–101 85

‘‘the ability of individuals in normal circumstances who are exposed to isolated and potentially highly disruptive eventsuch as the death of a close relation or a violent or a life-threatening situation to maintain relatively stable, healthy levelsof psychological functioning . . . as well as the capacity for generative experiences and positive emotions. (pp. 20–21).’’

As one of the leading scholars in this area, Bonanno and colleagues have documented evidence that, on average, peopleare remarkably resilience to bereavement (Bonanno et al., 2002; Bonanno, Moskowitz, Papa, & Folkman, 2005) and, morerecently, to direct exposure of the September 11th, terrorist attack in 2001 (Bonanno, Rennicke, & Dekel, 2005; Bonanno,Galea, Buccoarelli, & Vlahov, 2006). Using a measure of posttraumatic stress disorder (PTSD) as the outcome variable of inter-est, they were able to show that although an average adult tends to experience some short-term drop in his emotional andphysical well-being following a traumatic event, his reactions to it tend to be relatively brief and usually do not impede hisability to function to a significant degree.

While studies by Bonanno and other scholars1 have shown that evidence of psychological resilience to traumatic events iscommon, little continues to be known about the heterogeneity and the determinants of the heterogeneity that forms the aver-age (Bonanno, 2005). For instance, such questions as ‘Why are certain individuals better than others at bouncing back from abad life event?’ and ‘Why are they initially hurt less by such a shock?’ continue to be imperfectly understood.

There is little theoretical work on psychological resilience. One of the more notable studies is the work by economistsGraham and Oswald (2010), in which psychological resilience is conceptualized as a by-product of how much stock of hedo-nic capital the individual has accumulated over the years. According to them, their definition of hedonic capital could include

‘‘social relationships with partners, friends and colleagues; health; self-esteem; status; and meaningful work. For somepeople, religious faith may also play a part. These things are stocks in that they rely on past inputs and are carried acrosstime periods. (p. 373).’’

What this implies is that our ability to cope with stress and adversity could have been accumulated early on in our child-hood, and if we have lost a lot of psychological resources while we were young, it is likely that we will grow up to be lesspsychologically resilient in the future.

The conceptual framework of Graham and Oswald’s model is arguably implicitly embedded in the household productionmodel developed by Becker (1981) and the technology of skill formation model developed by Cunha and Heckman (2007).According to these models, early family and school inputs play a very important role in determining outcomes in adulthood,including cognitive and non-cognitive skills, earnings, employment, marital stability, health, and satisfaction with life over-all.2 It is thus possible that individual’s ability to cope with stress and adversity is also determined in the same way as cognitiveand non-cognitive skills. What this implies is that we may be able to use measures of family and school inputs, as well as indi-cators of different experiences in childhood and adolescence, to predict the extent of psychological resilience in adulthood.

The empirical evidence in this area comes primarily from studies in the psychology literature.3 Findings range from theexperiences of adversity in childhood – e.g. from having been sexually abused or a victim of violent crime – being detrimental(Compas, Howell, Phares, Williams, & Guinta, 1989; Enoch, 2011; Flores, Cicchetti, & Rogosch, 2005) to its having a positiveinfluence on later resilience (DuMont, Widom, & Czaja, 2007; Flynn & Biro, 1998; Flynn, Ghazal, Legault, Vandermeulen, &Petrick, 2004). A number of studies, however, have reported having a good parent-child relationship to be one of the single mostimportant moderators of adverse life events on child resilience and resilience in adulthood (DuMont et al., 2007; Masten, Best, &Garmezy, 1990). In addition to this, Brennan, Brocque, and Hammen (2003) have found that low levels of parental psychologicalmanipulation, high levels of maternal warmth, and low levels of maternal overinvolvement all interacted with maternal depres-sion to predict resilience outcomes in youth. Other studies have found objective measures of life successes which may havebeen accumulated since childhood such as income and wealth to have a protective effect on people’s risk of developing apost-traumatic stress disorder (PTSD) following a negative life event such as the September 11th attacks and a death of lovedones (Bonanno, Galea, Buccoarelli, & Vlahov, 2007; Galatzer-Levy & Bonanno, 2012; Hobfoll, 2002).

The current paper contributes to the economics literature on psychological resilience by asking a relatively unexploredquestion: Can certain childhood characteristics predict the extent of the psychological cost to one of the most important eco-nomic shocks, i.e. job loss? We know from previous studies in the well-being literature that unemployment induces, on aver-age, one of the largest negative effects on individual’s mental well-being and life satisfaction (e.g. Clark & Oswald, 1994;Darity & Goldsmith, 1996; Powdthavee, 2012; Winkelmann & Winkelmann, 1998). When economists have started lookingmore closely at the data, however, they found the negative effect of unemployment on individual’s psychological well-beingis significantly smaller in areas of high unemployment rates (Clark, 2003; Clark, Knabe, & Rätzel, 2009; Powdthavee, 2007).What this implies is that other people’s unemployment has a stress-buffering effect on individual’s own unemployment.Although this finding of unemployment as a social norm supports the idea that people from different contexts and back-grounds may react emotionally differently to becoming unemployed, the empirical evidence is small and the nature ofthe phenomenon continues to be imperfectly understood in the economics literature.

1 See, e.g. Freedy, Shaw, Jarrel, and Master (1992), Galea et al. (2002) and Boscarino et al. (2004).2 For empirical evidence of this, see, e.g. Todd and Wolpin (2007), Cunha, Heckman, and Schennach (2010) and Frijters, Johnston, and Shields (2014).3 For a few exceptional studies in the economics literature, see the work by Clark and Lelkes (2005) and Boyce and Wood (2011) who found religion and

personality traits help buffer shocks on individual’s well-being.

86 N. Powdthavee / Journal of Economic Psychology 45 (2014) 84–101

In an attempt to fill part of this research void, the current study utilizes a unique, nationally-representative British longitu-dinal data set that tracked almost 3000 adolescents (aged 11–15) into adulthood to explore which selected childhood charac-teristics predict the extent of people’s psychological reaction – or what looks like a degree of psychological resilience – to futureunemployment shocks. We base our selection of the childhood variables on what had been found in the psychology literature tobe important predictors of adult resilience (Bonanno et al., 2007; Masten et al., 1990), as well as what are available in the Britishyouth data. We then estimate one of the first micro-econometric mental well-being and life satisfaction regression equations inwhich an onset of unemployment in adulthood is interacted with an array of childhood characteristics of the survey respondent.We show that, on average, having a poor relationship with father as an adolescent predicts a significantly larger psychologicalreaction to unemployment shock in adulthood. This is robust even when we also allowed for the interactions between unem-ployment and factors that may be correlated with parent-child interaction, e.g. paternal mental well-being, paternal unemploy-ment, and family income in childhood. By contrast, having an unemployed mother during the child’s adolescent years appearsto have buffered the negative unemployment effect experienced by the child in the future. There are also gender differences toour findings, with different childhood characteristics having different predictive power on male’s and female’s psychologicalresponses to joblessness. Although little causal inferences could be made from these estimates, and further research is neededat this stage, the potential implications of our results are discussed.

2. Data

Our data set comes from the youth sample and the main adult sample of the British Household Panel Survey (BHPS). TheBHPS is nationally representative of British households, contains over 14,000 adult individuals aged 16 and over, and hasbeen conducted between September and Christmas each year since 1991 (Taylor, Brice, Buck, & Prentice-Lane, 2002). Theyouth sample – all children aged between 11 and 15 years – was first introduced to the BHPS in 1994 (Wave 4), and currentlyconsists of around 800–1400 person-year observations of youths in any given survey wave. These adolescents continue to beinterviewed as part of the youth survey until they turn 16 before entering the main adult sample.

We make use of both the youth sample and the adult samples in our analysis. We use thirteen waves of the youth surveys,i.e. Waves 4–17 (or years 1994–2007). However, it should be noted that the maximum number of times that each individualcould be interviewed in the youth survey is five, i.e. from aged 11–15. What this means is that for youth i who was eleven yearsold in Wave 4, his last wave in the youth survey will be in Wave 8 when he turned fifteen. When generating our main data set,we first take the within-person averages of everyone in our youth data. These within-person averages are then matched usingthe PID variable to the same person in the adult data. In other words, each person in the main BHPS survey, if they were inter-viewed as youths in the past, would have been matched with the average responses that they gave in the youth data set. Basedon previous findings in the psychology literature (e.g. Bonanno, 2004; Hobfoll, 2002; Masten et al., 1990), the following child-hood characteristics – which range from measures of parent-child relationships to the mental distress of the parents – are cho-sen as potential risk or protective factors to the negative psychological effect of unemployment in adulthood:

(i) The number of child’s close friends.(ii) The frequency of child fighting with someone at school.

(iii) The frequency of child having arguments with father.(iv) The frequency of child having arguments with mother.(v) The frequency of child talking to dad about things that matter.

(vi) The frequency of child talking to mum about things that matter.(vii) Father’s mental distress level.

(viii) Mother’s mental distress level.(ix) Father’s unemployment.(x) Mother’s unemployment.

(xi) Log of real household income per annum.

The number of child’s close friend is taken from the child’s response to the question, ‘‘How many close friends do you have –friends you could talk to if you were in some kind of trouble?’’ with answers ranging from 1 to 11 close friends.

The frequency of child fighting with someone at school comes from asking the child, ‘‘How often in the past month had youhad a fight with someone that involved physical violence, such as hitting, punching, or kicking?’’ Possible responses are ‘‘None’’,‘‘Once’’, ‘‘2 – 5 times’’, ‘‘6 – 9 times’’, and ‘‘10 or more’’.

The frequencies of having arguments with father/mum are based on the following question, ‘‘Most children have occasionalquarrels with their parents. How often do you quarrel with your father/mother?’’ Possible answers include ‘‘Hardly ever’’, ‘‘Lessthan once a week’’, ‘‘More than once a week’’, and ‘‘Most days’’.

The frequencies of child talking to dad/mum about things that matter come from asking children, ‘‘How often do you talkto your father/mother, about things that matter to you?’’, with possible answers ranging from ‘‘Hardly ever’’, ‘‘Less than once aweek’’, ‘‘More than once a week’’, and ‘‘Most days’’.

Parental mental distress is derived from the General Health Questionnaire (GHQ-12) score, which was self-reported bythe parents of the child. Many medical scholars and other researchers consider the scale to be a good proxy for mental stress


and strain (e.g. Guthrie et al., 1998). Individuals indicate on a 4-point scale from 1 (no more than usual) to 4 (much morethan usual) how often over the past few weeks they had lost sleep over worry, felt constantly under strain, felt they couldnot overcome difficulties, been feeling unhappy and depressed, been losing confidence, and been feeling like a worthless per-son. Individuals were also asked to indicate on a 4-point scale from 1 (better than usual) to 4 (much less than usual) on howoften over the past few weeks that had felt that they were playing a useful part in things, felt capable of making decisions,been able to enjoy day-to-day activities, been able to concentrate, been able to face up to problems, and been feeling rea-sonably happy. We utilize the Caseness score of GHQ, in which the number of times the person places himself or herselfin the fairly stressed or highly stressed category are added up together to form a total score. This is the BHPS variableHLGHQ2, with a scale running from 0 (lowest mental distress) to 12 (highest mental distress).

Finally, information on parental unemployment is drawn from a question in the main BHPS survey that asks all adultsabout their current labour force status, whilst real household income per annum during the adolescent years is taken fromthe derived household income variable in the BHPS divided by yearly consumer price index (CPI) which was then matched tothe youth survey.

The correlation matrix of the eleven standardized average childhood characteristics is reported in Table A1 in Appendix A.Looking across the columns of Table A1, the highest correlation is 0.55, which comes from the relationship between the fre-quency of child having arguments with father and the frequency of child having arguments with mother. This implies thatthere is unlikely to be a problem of multicollinearity across the different childhood characteristics.

Our main sample consists of all adults aged 16 and over who had previously been interviewed in the youth survey. Forexample, adolescents who were 15 years old in Wave 4 of the BHPS would have been around 29 when they were interviewedas adults in Wave 18. We focus our attention only on children where both parents were present and traceable in the youthpanel, which enables us to estimate the effects of both paternal and maternal unemployment on adult resilience in the sameregression equation. 4 This produces an unbalanced panel of 8779 observations (1928 individuals). Of those, 4479 observations(952 individuals) are females, and 4300 observations (976 individuals) are males.

Information on individual’s unemployment is drawn from a question in the main BHPS survey that asks all adults abouttheir current labour force status. The unemployment rate in the sample stands at around 7% (N = 586). Of those, 239 obser-vations of females are unemployed and 349 observations of males are unemployed. It is worth noting that there are poten-tially many negative life events that we could use to test our hypotheses; for example, bereavement, disability,unemployment, etc. However, given the relatively young age of our adult sample (aged between 16 and 29 years old), itseems that the most appropriate life shock – in terms of observed cases – is the unemployment shock on individuals.

The measures of individual’s well-being, which we use as the dependent variables, are mental well-being (or the inversedscores of the ‘‘Caseness’’ GHQ-12 described earlier, with a scale now running from 0 (lowest mental well-being) to 12 (high-est mental well-being) and life satisfaction. Responses to the life satisfaction question are elicited using the following ques-tion: ‘‘All things considered, how satisfied or dissatisfied are you with your life overall using a 1–7 scale? 1 = very dissatisfied, . . .,7 = very satisfied’’. The life satisfaction measure has been shown in the literature to represent a measure of cognitive well-being as opposed to measures of affect well-being such as the GHQ-12 (Diener, Emmons, Larsen, & Griffin, 1985).

3. Empirical strategy

To fix ideas, let us assume that there exists a reported well-being function of the following form

4 Thesurveys

r ¼ hðf ðu;u � c; x; tÞ þ eÞ; ð1Þ
where r denotes some self-reported number or level collected in the survey. The f(. . .) function is the individual’s true well-beingand is observable only to the person asked; h(�) is a non-differentiable function relating actual to reported well-being; u isunemployment; c is a set of childhood characteristics; x represents a set of socio-economic characteristics other than the unem-ployment status of the respondent; t is time trend; and e is an error term that subsumes the respondent’s inability to commu-nicate accurately his or her well-being levels. Here, we assume that own unemployment has a detrimental effect on individual’swell-being, but this negative effect is either moderated or exacerbated by what the respondent has gone through for the mostpart of his or her childhood or adolescence. Our main empirical model, which is a counterpart of (1), is then given by
wit ¼ aþ b0uit þX11

s¼1

�Cisbs þX11

s¼1

ðuit � �CisÞcs þ x0itkþ v i þ eit ; ð2Þ

where i = 1, 2, . . ., n; t = 16, . . ., 29 years old; vi is the unobserved individual fixed effects, and vit is the random-error term. Inan attempt to capture as much information about the respondent’s time in the youth panel as possible, �C here represents aset of within-person averages of childhood variables taken from the unbalanced youth survey, i.e. aged 11–15. To aid inter-pretation of the results in our fully interacted model, the within-person averages of childhood characteristics, which aretime-invariant, are standardized across the sample to have a mean of zero and a standard deviation of one. What thisimplies is that we can interpret the coefficient on unemployment, b0, as the estimated well-being effect of unemploymenton respondents whose standardized childhood averages are equal to zero, and b0 þ k1 as the estimated well-being effect of

sample thus consists mostly of married/cohabiting parents, with a handful of separated and divorced parents who continued to participate in thedespite living in different households.


unemployment on respondents whose one of the standardized childhood averages is one standard deviation above themean.

All of our estimation is done using either random effects or fixed effects linear model with cluster-robust standard errors(clustered at the individual level) (Cameron & Miller, 2013), although similar conclusions can be reached when a ConditionalLogit estimator was used to estimate a model with a dichotomous well-being measure as an outcome variable (Ferrer-i-Carbonell & Frijters, 2004; see also Clark, Georgellis, & Sanfey, 2001).

4. Results

Which childhood characteristics predict the extent of psychological reaction to unemployment in adulthood? Table 1 pro-vides a first pass at this question. By comparing the raw means of mental well-being of the employed and the unemployedand their differences by childhood characteristics, we were able to arrive at three different conclusions.

The first, which is a standard finding, is that mental well-being and life satisfaction are generally lower for the unem-ployed than for the employed.

Table 1Raw data summary of mental well-being and life satisfaction of the employed and the unemployed by childhood characteristics.

EL UL EH UH DD

Panel A: Mental well-beingNumber of close friends 10.28 9.41 10.34 9.54 0.15

(0.05) (0.15) (0.07) (0.17)Frequency of fighting with someone 10.31 9.52 10.28 9.39 �0.09

(0.05) (0.15) (0.06) (0.16)Frequency of having arguments with dad 10.54 9.77 10.05 9.19 �0.07

(0.05) (0.15) (0.06) (0.16)Frequency of having arguments with mum 10.52 9.72 10.01 9.05 0.02

(0.05) (0.14) (9.49) (0.19)Frequency of talking to dad about things that matter 10.32 9.32 10.28 9.66 0.38

(0.05) (0.15) (0.06) (0.16)Frequency of talking to mum about things that matter 10.38 9.52 10.22 9.38 0.02

(0.05) (0.15) (0.06) (0.17)Father’s mental distress (GHQ-12) 10.49 9.69 10.07 9.31 0.09

(0.05) (0.17) (0.06) (0.15)Mother’s mental distress (GHQ-12) 10.49 9.80 10.03 9.12 �0.22

(0.05) (0.15) (0.07) (0.17)Father’s unemployment 10.32 9.23 10.24 9.77 0.63**

(0.04) (0.15) (0.08) (0.16)Mother’s unemployment 10.29 9.39 10.36 9.84 0.38

(0.04) (0.12) (0.14) (0.27)Log of real household income per annum 10.31 9.57 10.30 9.29 �0.19

(0.06) (0.14) (0.05) (0.18)

Panel B: Life satisfactionNumber of close friends 5.26 4.77 5.35 4.96 0.11

(0.02) (0.07) (0.03) (0.08)Frequency of fighting with someone 5.35 4.93 5.20 4.76 �0.06

(0.02) (0.07) (0.03) (0.08)Frequency of having arguments with dad 5.44 5.05 5.15 4.68 �0.09

(0.02) (0.07) (0.03) (0.08)Frequency of having arguments with mum 5.42 5.00 5.13 4.61 �0.14

(0.02) (0.07) (0.03) (0.09)Frequency of talking to dad about things that matter 5.26 4.78 5.33 4.96 0.12

(0.02) (0.07) (0.02) (0.08)Frequency of talking to mum about things that matter 5.23 4.75 5.36 4.99 0.12

(0.02) (0.07) (0.02) (0.08)Father’s mental distress (GHQ-12) 5.38 4.99 5.19 4.76 �0.04

(0.02) (0.08) (0.03) (0.07)Mother’s mental distress (GHQ-12) 5.35 4.91 5.21 4.79 0.01

(0.02) (0.07) (0.03) (0.08)Father’s unemployment 5.31 4.76 5.24 4.97 0.27**

(0.02) (0.07) (0.04) (0.08)Mother’s unemployment 5.29 4.84 5.38 4.91 �0.02

(0.02) (0.06) (0.06) (0.14)Log of real household income per annum 5.25 4.84 5.34 4.86 �0.09

(0.03) (0.07) (0.02) (0.08)

Note: *<10%. Mental well-being = inversed GHQ-12, with 0 = worst mental well-being, . . ., 12 = best mental well-being. Life satisfaction is on a 7-point scale(1 = very dissatisfied, . . ., 7 = very satisfied). EL = Employed individuals with lower than average level of childhood outcome j; EH = Employed individuals withthe average or higher level of childhood outcome j; UL = Unemployed individuals with lower than average level of childhood outcome j; UH = Unemployedindividuals with the average or higher level of childhood outcome j. DD = difference-in-difference, i.e., (UH� EH)� (UL� EL). Standard errors are in parentheses.

** <5%.


Secondly, within each employment category, we can see that people who had more close friends during adolescence, whotalked to their parents about the things that matter more often, whose parents spent more time in unemployment while theywere growing up, and who had higher household income as children were more likely than others to have better mentalwell-being and higher levels of life satisfaction as adults. By contrast, people who during their early teenage years had spenta significant amount of time fighting other people, arguing with their parents, and whose parents had reported lower mentalwell-being were more likely than others to have worse mental well-being and lower life satisfaction as adults.

Finally, and more importantly, we can see from the difference-in-difference at the cross-section that the well-being gapsbetween the unemployed and the employed are larger for individuals who, as adolescents, had fought with someone or hadspent time arguing with their parents relatively more often compared to the average. By contrast, we can also see that thewell-being gaps between the employed and the unemployed are smaller for individuals who had relatively higher numbersof close friends, who had talked to their father about the things that matter more often, or whose parents had spent moretime in unemployment during their early teenage years compared to the average. Nevertheless, we are unable to reject thenull hypothesis that the cross-sectional difference-in-difference between the high group and the low group is equal to zerofor almost every childhood characteristics.

Table 2 moves on to present our first econometric evidence. It reports the estimates for the random effects (RE) and fixedeffects (FE) micro-econometric models. The dependent variables are mental well-being (measured cardinally on the 0–12 scale)and life satisfaction (measured cardinally on the 1–7 scale).5 Exogenous control variables include age, age-squared, gender (in theRE estimation), year dummies, and regional dummies. Other socio-economic control variables include employment status (otherthan unemployment), marital status, income, education dummies, health dummies, number of children, and homeownership sta-tus. We also include the Big Five personality traits (extraversion, agreeableness, openness, neuroticism, and conscientiousness), aswell as their interactions with the unemployment variable, to allow for the possibility that the well-being effects of unemploymentvary across people of different personality types rather than by different childhood characteristics.6

Controlling only for the exogenous variables, we can see from the RE estimates in Columns 1 and 5 that unemploymententers both well-being equations in a negative and statistically significant manner. The estimated coefficient on unemploy-ment is �0.635 [S.E. = 0.138] in the mental well-being equation and �0.330 [S.E. = 0.059] in the life satisfaction equation,which is consistent with the previous findings in this area (Clark & Oswald, 1994; Winkelmann & Winkelmann, 1998;Clark, 2003). The estimated main effects of the standardized average childhood characteristics are negative and statisticallysignificant at conventional levels in both well-being regression equations for frequency of fighting with someone, frequencyof having arguments with father, frequency of having arguments with mother, father’s mental distress, and mother’s mentaldistress. Individuals who, as adolescents, had relatively more close friends, talked to their father about things that mattermore often, and had more income than the average tend to also be more satisfied with life as adults. It is interesting to notethat the main effects of father’s unemployment are positive and statistically significant in both sets of well-being equations.However, this may have been due to the fact that both father’s mental well-being and family incomes – both of which arehighly correlated with father’s unemployment – are also included in the regression equations.

Turning to the interaction effects, only the coefficients on ‘‘Unemployment� Frequency of having arguments with father’’ and‘‘Unemployment�Mother’s unemployment’’ are statistically significant at conventional levels in the mental well-being equation,although with opposite signs. Conditioning on being unemployed, a one standard deviation increase in the frequency of havingarguments with father is associated with a further drop in the unemployed’s mental well-being by approximately 0.28-point.On the other hand, a one standard deviation increase in the maternal unemployment variable is associated with an increase inthe unemployed’s mental well-being by approximately 0.27-point. The estimated coefficients are robust even when we accountfor other interaction effects between unemployment and factors such as paternal mental well-being, paternal unemployment,and household income during adolescent years. Nevertheless, the same cannot be said about the estimates in the life satisfac-tion equation, as none of the interaction terms are statistically significantly different from zero.

The RE estimates are, nevertheless, subject to potential bias from the presence of unobserved individual fixed effects, vi, asspecified in Eq. (2). For example, there may be time-invariant characteristics other than personality traits that jointly deter-mine one’s likelihood of having a bad parent-child interaction as an adolescent and the ability to withstand and adapt toeconomic shocks as an adult. This may include, e.g. individual and family characteristics that do not vary over time. Weaddress this issue by re-estimating Eq. (2) using the FE estimator and report the results in Columns 2, 3, 6, and 7 of Table 2.

Quantitatively similar results are obtained when we take individual fixed effects into account. Columns 2 and 6 producedthe following results for, say, an unemployed person who was one standard deviation above the mean in the ‘‘Frequency ofhaving arguments with father’’ variable. For these individuals, the estimated total effect of an unemployment shock on indi-vidual’s mental well-being is around 50% larger in size than the main unemployment effect at �0.603 � 0.292 = �0.895[S.E. = 0.251]. By contrast, the negative unemployment effect on mental well-being is almost completely offset for individualswho were one standard deviation above the mean in the ‘‘Mother’s unemployment’’; the estimated total unemployment effectfor these groups of individuals is negative though statistically insignificantly different from zero at �0.602 + 0.424 = �0.178[S.E. = 0.256]. It is worth noting here that the results are also robust to controlling for the interactions between unemploy-

5 All the paper’s results can be replicated with ordered estimators.6 These additional stock-like variables, which were obtained in Wave 15 of the BHPS, are entered with the same values across all waves for each respondent.

For example, if respondent i has responded with a score of 14 on the extraversion scale in Wave 15, her extraversion score will be 14 in every other wave whileshe is still in the panel.

Table 2Mental well-being and life satisfaction regression equations with interaction effects between average childhood characteristics and unemployment.

Variables Mental well-being Life satisfaction

All All All Aged 21+ All All All Aged 21+RE FE FE FE RE FE FE FE

Unemployment �0.635*** �0.603*** �0.765*** �0.345 �0.330*** �0.271*** �0.258** 0.106[0.138] [0.153] [0.238] [0.598] [0.059] [0.063] [0.114] [0.208]

Standardized average childhood characteristics (aged 11–15 years old)Number of close friends �0.037 0.066**

[0.054] [0.029]Frequency of fighting with someone �0.113** �0.073***

[0.053] [0.025]Frequency of having arguments with father �0.131** �0.048*

[0.059] [0.026]Frequency of having arguments with mother �0.137** �0.088***

[0.064] [0.028]Frequency of talking to dad about things that matter 0.038 0.106***

[0.057] [0.027]Frequency of talking to mum about things that matter �0.028 0.028

[0.060] [0.029]Father’s mental distress (GHQ-12) �0.153*** �0.078***

[0.051] [0.023]Mother’s mental distress (GHQ-12) �0.233*** �0.111***

[0.052] [0.024]Father’s unemployment 0.116*** 0.051**

[0.042] [0.024]Mother’s unemployment �0.027 �0.034

[0.062] [0.032]Log of real household income per annum 0.0250 0.085***

[0.054] [0.025]

Standardized average childhood characteristics (aged 11–15 years old) interacted with unemployment in adulthoodNumber of close friends �0.000 �0.028 �0.046 �0.094 0.055 0.043 0.048 0.288*

[0.151] [0.153] [0.164] [0.404] [0.068] [0.070] [0.070] [0.147]Frequency of fighting with someone �0.125 �0.199 �0.248 �0.300 �0.083 �0.061 �0.079 �0.003

[0.148] [0.161] [0.157] [0.265] [0.061] [0.064] [0.063] [0.134]Frequency of having arguments with father �0.278* �0.292 �0.350* �0.606 �0.063 �0.100 �0.116* �0.389***

[0.169] [0.203] [0.204] [0.374] [0.065] [0.069] [0.070] [0.130]Frequency of having arguments with mother 0.170 0.142 0.240 0.145 �0.009 0.015 0.051 0.235*

[0.178] [0.214] [0.218] [0.383] [0.073] [0.080] [0.081] [0.140]Frequency of talking to dad about things that matter 0.098 0.080 0.134 �0.119 0.029 0.006 0.028 0.080

[0.159] [0.183] [0.181] [0.435] [0.067] [0.069] [0.071] [0.163]Frequency of talking to mum about things that matter 0.032 �0.008 �0.085 �0.608 0.037 0.032 0.005 �0.070

[0.167] [0.187] [0.183] [0.407] [0.066] [0.068] [0.068] [0.140]Father’s mental distress (GHQ-12) 0.006 0.045 0.028 �0.070 �0.021 0.004 0.005 �0.111

[0.131] [0.137] [0.131] [0.254] [0.057] [0.059] [0.058] [0.101]Mother’s mental distress (GHQ-12) �0.065 �0.028 �0.087 �0.831* �0.010 �0.037 �0.055 �0.230

[0.148] [0.177] [0.175] [0.442] [0.060] [0.064] [0.062] [0.208]Father’s unemployment 0.040 �0.031 �0.076 �0.657* 0.016 0.003 0.007 �0.154

[0.117] [0.129] [0.123] [0.339] [0.068] [0.078] [0.073] [0.159]Mother’s unemployment 0.271** 0.424** 0.427** 0.845** 0.031 0.066 0.067 0.260*

[0.135] [0.197] [0.205] [0.385] [0.065] [0.066] [0.065] [0.148]

90N

.Powdthavee

/Journalof

Economic

Psychology45

(2014)84–

101

Log of real household income per annum 0.143 0.199 0.167 �0.554 0.035 0.053 0.056 �0.209[0.184] [0.209] [0.201] [0.403] [0.072] [0.076] [0.074] [0.164]

Exogenous control variables Yes Yes Yes Yes Yes Yes Yes YesSocio-economic control variables No No Yes Yes No No Yes YesObservations 8779 8779 8658 2230 8155 8155 7996 2170Number of individuals 1928 1928 1917 659 1918 1918 1893 658

Note: RE = random effects. FE = fixed effects. Exogenous control variables include age, age-squared, gender (in RE estimation), year dummies, and regional dummies. Socio-economic control variables include self-employment, full-time student, disabled, other non-labour activities, marital statuses, education dummies, log of real household income per annum, subjective health statuses, homeownership statuses, andinteraction effects between unemployment and the Big Five personality traits (extraversion, agreeableness, openness, neuroticism, and conscientiousness). Columns 1–3 and 5–7 include all respondents e.g. aged16–29. Columns 4 and 8 include only individuals who have likely left home e.g. aged 21–29.Cluster-robust standard errors (clustered at the individual level) are in parentheses.

* <10%.** <5%.

*** <1%.

N.Pow

dthavee/Journal

ofEconom

icPsychology

45(2014)

84–101

91

Fig. 1. The estimated effects of unemployment on mental well-being by childhood characteristics for all respondents (aged 16–29), regression-corrected.Note: 2-standard-error bands (90% C.I.) are reported: one s.e. above and one below. The vertical line represents the estimated effect of unemployment whenall childhood characteristics are at their means, i.e. zero. The estimates are based on Column 3 of Table 2’s specification.


ment and the Big Five personality traits. Note also that the main effects of the standardized average childhood characteristicswere naturally dropped from the FE regressions as these do not change over time.

The results hardly changed following the inclusion of socio-economic control variables in Columns 3 and 6 of Table 2.7

However, the interaction term ‘‘Unemployment � Frequency of having arguments with father’’ is now negative and statisticallysignificant at the 10% level in the life satisfaction equation. The estimated heterogeneous effects of unemployment by differentchildhood characteristics on mental well-being and life satisfaction are better illustrated in Figs. 1 and 2.

The extra drop in mental well-being for individuals who had frequent arguments with their father as adolescents is quan-titatively important as well as statistically significant. Take, for example, the averaged unemployed person and the unem-ployed person whose ‘‘Frequency of having arguments with father’’ between the ages of 11 and 15 is one standarddeviation higher than the average. The gap in mental well-being between these two groups of individuals is �0.350, witha statistically significant standard error of 0.204. Given the distribution of the mental well-being score, this is a large effect.It can completely offset the positive effect on mental well-being from becoming an outright homeowner or getting an A-levelqualification; it is approximately equal to 25% of the negative effect brought about by disability; and it is slightly larger thanthe negative effect brought about by having one extra child.

As a robustness check, Columns 4 and 8 re-estimate FE equations on a restricted sample of individuals aged 21 and over,i.e. those who are likely to have left home, thus making them less prone than other younger children in the sample to becontinually exposed to the same household environments such as argumentative parents and/or caring parents. Since themajority of our adult sample is made up of people in their late teens, running FE on this restricted sample means that welose around 6400 (or 75%) of our original observations. Despite this, we continue to find, even in this significantly smallersample, statistically significant interaction effects for both ‘‘Frequency of having arguments with father’’ and ‘‘Mother’s unem-ployment’’ at conventional levels.

Table 3 separates the data by gender of the respondent. With the exception of male’s life satisfaction, unemployment con-tinues to be associated negatively and statistically significantly with individual’s well-being in general. The estimated coef-ficient on the interaction term ‘‘Unemployment � Frequency of having arguments with father’’ remains negative throughout,although it is now only marginally significant in the female’s mental well-being equation. Having had a relatively more dis-ruptive time at school – i.e. recording a higher than average ‘‘Frequency of fighting with someone’’ – now predicts significantlylower levels of psychological resilience to job loss for men. Additionally, we also find women whose mother had spent moretime in unemployment compared to the average when they were between 11 and 15 years old to be hurt less psychologicallyfrom becoming unemployed. Hence, what Table 3’s results are implying is that the structure of a psychological resilienceequation may be different across gender groups.

One question of interest is whether we can still obtain similar results with childhood characteristics that had been col-lected at different stages of child development? According to many studies in the household production function of humancapital literature, the returns to family inputs on early childhood investments are typically larger than those on investmentsin late childhood and adolescents (Carneiro & Heckman, 2003; Cunha & Heckman, 2008; Heckman, 2000). Other types ofinputs such as school and peers may also affect children differently at different ages. To test this, we replace the standardizedaverage childhood characteristics between eleven and fifteen in Eq. (2) by (a) the standardized childhood characteristics

7 See Table A1 in Appendix A for the rest of the estimates in Columns 3 and 7 of Table 2.

Fig. 2. The estimated effects of unemployment on life satisfaction by childhood characteristics for all respondents (aged 16–29), regression-corrected. Note:2-standard-error bands (90% C.I.) are reported: one s.e. above and one below. The vertical line represents the estimated effect of unemployment when allchildhood characteristics are at their means, i.e. zero. The estimates are based on Column 6 of Table 2’s specification.

Table 3Mental well-being and life satisfaction regression equations by gender for all respondents (aged 16–29), fixed effects regressions.

Variables Men Women

Mental well-being Life satisfaction Mental well-being Life satisfaction

Unemployment �0.928** �0.015 �0.861** �0.368**

[0.437] [0.160] [0.337] [0.161]

Standardized average childhood characteristics (aged 11–15 years old) interacted with unemployment in adulthoodNumber of close friends 0.008 �0.007 �0.065 0.123

[0.203] [0.080] [0.261] [0.105]Frequency of fighting with someone �0.352** �0.152** �0.460 �0.001

[0.170] [0.077] [0.312] [0.136]Frequency of having arguments with father �0.084 �0.120 �0.406 �0.017

[0.300] [0.119] [0.274] [0.098]Frequency of having arguments with mother 0.252 �0.024 0.330 0.121

[0.274] [0.118] [0.301] [0.111]Frequency of talking to dad about things that matter 0.207 0.204** 0.093 �0.049

[0.239] [0.101] [0.268] [0.124]Frequency of talking to mum about things that matter �0.151 �0.109 �0.051 0.120

[0.245] [0.104] [0.290] [0.102]Father’s mental distress (GHQ-12) �0.212 �0.104 0.325 0.080

[0.192] [0.065] [0.200] [0.096]Mother’s mental distress (GHQ-12) 0.093 0.031 �0.193 �0.077

[0.192] [0.077] [0.318] [0.099]Father’s unemployment �0.042 0.073 �0.068 0.009

[0.179] [0.091] [0.218] [0.138]Mother’s unemployment 0.037 �0.212* 0.684** 0.256***

[0.264] [0.110] [0.275] [0.082]Log of real household income per annum 0.205 0.032 0.202 0.114

[0.272] [0.099] [0.321] [0.109]

Exogenous control variables Yes Yes Yes YesSocio-economic control variables Yes Yes Yes YesObservations 4240 3906 4418 4090Number of individuals 972 957 945 936

Note: Cluster-robust standard errors (clustered at the individual level) are in parentheses. Same set of control variables as in Table 2.* <10%.

** <5%.*** <1%.


measured when the child was eleven in one set of equations, and (b) the standardized childhood characteristics measuredwhen the child was fifteen in another set of equations. We then re-estimate these equations and report the new results inTables A3 and A4 in Appendix A.


On the whole, we find evidence that the extent of negative psychological response to unemployment in adulthood may beexplained by more childhood characteristics measured at eleven than characteristics measured at fifteen. For example, theinteraction term ‘‘Unemployment � Frequency of having arguments with father’’ is negative and statistically significant only inequations where the variable at eleven is used. We also find evidence of a significantly higher level of negative psychologicalreaction to job loss in terms of mental well-being for adults whose mother scored very low on the mental well-being scalewhen they were eleven years old. Having had a father to talk to about things that matter at this relatively early age predictsfuture resilience to unemployment, especially for men. Nevertheless, there are a few characteristics at fifteen that appear tomatter more at predicting what resembles psychological resilience in adulthood compared to those measured at eleven,including, for example, household income in the full sample (see the mental well-being estimates) and number of closefriends in the male sample (see the life satisfaction estimates).

What are the possible mechanisms that could be used to explain why children of certain characteristics grew up to be moreor less psychologically sensitive to unemployment shock? More specifically, what might explain why children from relativelymore argumentative households might be affected more psychologically by an unemployment shock, on average? One possibleexplanation for this is that children from these households may go on to receive less education and consequently will be morevulnerable to the psychic cost of joblessness because of the lower future employment opportunities that they have to face. How-ever, though not reported here, our estimates on the childhood interaction effects remain statistically robust even in regressionswhere interactions between unemployment and different levels of education are allowed for, thus suggesting that the early con-frontations with their father may have had a long-lasting impact on the unobserved psychological nature of the individuals thathave not been accurately captured using the Big Five inventory. This may include – but not limited to – self-esteem and locus ofcontrol, the two factors that could also act as psychological buffers to an unemployment shock. Other explanations are also pos-sible – e.g. children who argued a lot with their parents may find it hard to make friends as adults and therefore are less likely tohave the kind of social supports they need to cope with becoming unemployed in adulthood.

What about the seemingly paradoxical result regarding the positive effect of maternal unemployment during adolescents onthe well-being of the unemployed, evidence that is much stronger for females than for males? One possible explanation for thisis the social norm effect of other people’s unemployment – i.e. the finding that unemployment hurts less psychologically whenthere is more of it around (e.g. Clark, 2003; Eggers, Gaddy, & Graham, 2006; Clark et al., 2009; Powdthavee, 2007). It is imag-inable then that maternal unemployment experienced during the child’s adolescent years may have had a long-lasting effect ontheir daughter’s attitudes toward their own future unemployment. Another explanation might be that unemployed mothers getto spend significantly more time with their children during their unemployment spells, which could have had a positive andlong-lasting impact on the stock of their children’s psychological resources. This is not inconsistent with the latest findings thatearly maternal unemployment when the child was relatively young is positively correlated with self-reported happiness of thechild (Powdthavee & Vernoit, 2013).8 Nevertheless, it should also be noted here that we do not observe the same positive exter-nality of paternal unemployment on sons’ emotional reactions to own unemployment in our data set.

More broadly, our results seem to suggest that children who generally had lower levels of psychological resources inchildhood tend to be less resilient to economic shocks in the future (Graham & Oswald, 2010). Yet, in the absence of random-ized childhood characteristics and experiences, we cannot rule out that the interpretation of our results might have nothingto do with early losses of psychological resources. It may be that children of certain attributes, e.g. those with poor parent-child relationships, are more likely than others to select themselves into unemployment in the first place. If these attributesalso happened to be correlated negatively with children’s well-being, then it is reasonable to assume that these children willalso grow up to be more fearful and anxious as adults (and thus are more negatively affected by job loss).

To shed some light on this issue, Table 4 divides the sample by employment status and checks whether people of certainchildhood characteristics select themselves more often into unemployment as adults. Looking across the columns, we cansee that the unemployed had, on average, a significantly more disruptive childhood compared to the employed. A typicalunemployed person had, for example, a lot more fights that took place at school, more distressed parents, and less householdincome as adolescents than the employed, and we can reject the null hypothesis of equal means at conventional confidencelevels. However, we cannot reject the null hypothesis that the employed and the unemployed are the same in terms of ‘‘Fre-quency of having arguments with father’’, which was one of the only three factors that significantly explained the extent ofpsychological resilience to job loss in our full sample analysis. Hence, one conclusion here is that people of certain childhoodcharacteristics may indeed select themselves into unemployment more often than others, but it is unlikely that this willexplain everything about the extent of psychological resilience to job loss in adulthood. At any rate, it is still advisable forreaders to treat our interpretations of the coefficients with care.

The next question of interest is whether adaptation to unemployment is slow and incomplete for certain groups of indi-viduals. We do this by expanding Eq. (2) to include leads and lags to unemployment – one-year lead and two-year lags to beprecise – as well as their interactions with different childhood characteristics. We then estimated this new equation usingthe FE estimator on a sample in which at least three years of mental well-being and life satisfaction are consecutively

8 To test this potential pathway, we estimated another specification that includes an interaction term ‘‘Unemployment �Mother’s non-labour status’’ as anadditional variable in the equation. Similar to the interaction term ‘‘Unemployment �Mother’s unemployment’’, the estimated coefficient on this interaction termis positive albeit statistically insignificant in both mental well-being and life satisfaction equations, thus implying that the positive externality from maternalunemployment might not be due entirely to the relatively more time unemployed mothers have for the child while he/she was growing up.

Table 4Selection into unemployment by childhood characteristics.

All Proportion of people employed Proportion of people unemployed

Number of close friends 0.006 0.072Frequency of fighting with someone 0.030 0.265 ***

Frequency of having arguments with father 0.035 0.062Frequency of having arguments with mother �0.002 0.016Frequency of talking to dad about things that matter �0.054 �0.200 *

Frequency of talking to mum about things that matter �0.063 �0.227 **

Father’s mental distress (GHQ-12) �0.057 0.174 ***

Mother’s mental distress (GHQ-12) �0.026 0.254 ***

Father’s unemployment �0.001 0.287 ***

Mother’s unemployment �0.051 0.286 ***

Log of real household income per annum �0.058 �0.407 ***

MenNumber of close friends 0.042 0.065Frequency of fighting with someone 0.066 0.262 *

Frequency of having arguments with father 0.012 0.113Frequency of having arguments with mother 0.007 0.082Frequency of talking to dad about things that matter �0.059 �0.184Frequency of talking to mum about things that matter �0.064 �0.172Father’s mental distress (GHQ-12) �0.040 0.128Mother’s mental distress (GHQ-12) �0.015 0.269 ***

Father’s unemployment 0.030 0.352 **



WomenNumber of close friends 0.006 0.132Frequency of fighting with someone �0.023 0.188 *

Frequency of having arguments with father 0.047 0.009Frequency of having arguments with mother �0.004 �0.023Frequency of talking to dad about things that matter �0.018 �0.207 *

Frequency of talking to mum about things that matter �0.003 �0.177Father’s mental distress (GHQ-12) �0.082 0.282 ***

Mother’s mental distress (GHQ-12) �0.039 0.240 **

Father’s unemployment �0.042 0.165 *



Note: Figures in each cell represent the proportions of people employed and unemployed by childhood variables. Asterisks are based on the t-test that thetwo means – E versus U – are statistically the same.

* <10%.** <5%.

*** <1%.


observed.9 Our empirical strategy here is similar to the one adopted by Clark et al. (2008), Frijters et al. (2011) and Powdthavee(2012). Given that the table produced a large number of coefficients, we present the geographical representations of the impliedwell-being effects of having spent three consecutive years in unemployment (from T to T + 2) only for two groups of individuals:those who were below the average level in the standardized average ‘‘Frequency of having arguments with father’’ variable andthose who were above it10 – and display them in Fig. 3a (mental well-being) and b (life satisfaction).

When we compare the dynamics of both well-being measures across these two groups of individuals, we can see thatthere is some evidence of a one-year anticipation effect to becoming unemployed at T for people who were above the averagelevel in the frequency of having arguments with father as adolescents. Although neither group has completely adapted tounemployment at T + 2 – a result that is generally consistent with the previous findings on adaptation to unemployment(Clark et al., 2008; Powdthavee, 2012), the estimated unemployment effect is, in most time periods, more negative forthe people who were above the average level in the standardized average ‘‘Frequency of having arguments with father’’ var-iable compared to the people who were below it.

We next ask whether the extent of people’s ability to cope with job loss varies significantly by unemployment types. Todo this, we follow the research of Kassenboehmer and Haisken-DeNew (2009) and split the unemployment shock into unem-ployment by redundancies and unemployment by other reasons, e.g. dismissed, left for health reason, or temporary jobended, etc. Assuming that redundancies – which make up to around 15% of the unemployed – represent exogenous (or invol-untary) changes in employment status, we re-estimated the FE specification of Eq. (2) on mental well-being and life

9 The random effects specification is available upon request.10 Since this is one of the only three variables (the other two being frequency of fighting with someone and maternal unemployment) that, when having been

interacted with the unemployment variable, is statistically significantly different from zero in the full sample regression.

(a) Mental well-being

(b) Life satisfaction

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

T-1 T T+1 T+2

Esti

mat

ed u

nem

plo

ymen

t eff

ect o

n

men

tal w

ell-

bei

ng

(in

vers

edG

HQ

-12

)

Below average in the frequency of having arguments with father

Above average in the frequency of having arguments with father

-1.6

-1.4

-1.2

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

T-1 T T+1 T+2

Esti

mat

ed u

nem

plo

ymen

t eff

ect o

n li

fe

sati

sfac

tion

Below average in the frequency of having arguments with father

Above average in the frequency of having arguments with father

Fig. 3. The estimated well-being effects before and after unemployment for people of low and high frequencies of having arguments with father duringadolescence, full sample. (a) Mental well-being. (b) Life satisfaction. Note: Year T is the year of unemployment. The individuals also remained inunemployment in T + 1 and T + 2. 2-standard-error bars (90% C.I.) are reported: one s.e. above and one below.

Table 5The estimated well-being effects of different types of unemployment by frequencies of having arguments with father during adolescence, full sample.

The implied well-being effects of unemployment Mental well-being Life satisfaction

Unemployed (other reasons) �0.618*** �0.294***

[0.232] [0.109]Unemployed (redundancies/layoffs) �0.999*** �0.135

[0.346] [0.125]Unemployed (other reasons) + 1 S.D. of frequency of having arguments with father �0.975*** �0.434***

[0.320] [0.125]Unemployed (redundancies/layoffs) + 1 S.D. of frequency of having arguments with father �1.507*** �0.160

[0.474] [0.139]

Note: Cluster-robust standard errors (clustered at the individual level) are in parentheses.*** <1%.


satisfaction, and report the implied well-being effects of unemployment in Table 5. Again, as in the previous table, we focusedonly on the interaction effects between unemployment and the ‘‘Frequency of having arguments with father’’ variable here.


We find people who became unemployed by redundancies to report, on average, significantly larger drops in mental well-being – but not in life satisfaction – compared to those who became unemployed for other reasons. However, we also findsome evidence that unemployment – redundancies or otherwise – depressed mental well-being significantly more for thosewho reported to have had relatively more frequent arguments with their father in the past. For example, while an exogenousunemployment shock is associated with a drop in mental well-being for an average person by approximately 1-point, thenegative effect is 50% larger at 1.507 [S.E. = 0.404] for individuals who were one standard deviation above the average inthe ‘‘Frequency of having arguments with father’’ variable. Finally, we are not able to find evidence to suggest that the extentof people’s ability to withstand job loss varies significantly by unemployment types, i.e. we cannot reject the null hypothesisthat the coefficients on ‘‘Unemployment (others) � Frequency of having arguments with father’’ and ‘‘Unemployment (redundan-cies) � Frequency of having arguments with father’’ are the same.

5. Conclusion

This paper uses data on mental well-being and life satisfaction to test theories of psychological resilience in economics. Itempirically investigates whether psychological resilience to economic shocks in adulthood can be predicted using variablesobtained in childhood. Consistent with the previous studies on the psychological effects of job loss, an unemployment shockis associated with a significant drop in both mental well-being and life satisfaction for the individuals. However, the drop inwell-being is predicted to be around 50% larger for people who were one standard deviation above the mean in the frequencyof having arguments with father between aged 11 and 15. Maternal unemployment, on the other hand, is found to be asso-ciated positively with the well-being of the unemployed. However, there are also many other factors such as having moreclose friends or father’s mental distress that do not seem to predict heterogeneity in the psychological response to job lossin adulthood at all.

This paper is not without limitations. One natural objection to our results is that the relationship between, say, having apoor father-child relationship during childhood and psychological resilience in adulthood is not causal. While our FE estima-tion may have taken care of the unobserved person-specific characteristics that made people argued more with their fatheras children and less psychologically resilient as adults, we still do not know what may have caused such a poor father-childrelationship to cultivate in the first place. Although this problem is not easily solved without a good identification strategy,we still believe that the ability to predict who will be more negatively affected by future economic shocks is one of the keycomponents to optimal policy design. What this means is that even if our indicator of having a poor father-child relationshipis nothing more than just a mirror reflection of something else that is unobserved about the child – e.g. time-persistent unob-served personality traits, then at least it serves its purpose as a device which can be used to gauge early a child’s ability tocope and adapt to negative shocks in the future.

Another important objection, which was raised by one of the referees, is that the observed moderation effects on the neg-ative psychological impact of unemployment may be due to short-run contemporaneous impacts of the selected childhoodcharacteristics as well as long-run lagged impacts, which has so far been our preferred interpretation of the estimated interac-tion effects. Given that the expected length of an unemployment spell as well as its effect on saving, consumption, and otherliving conditions is a function of contemporaneous factors, and that these contemporaneous factors are themselves relatedto the selected childhood characteristics, one might even conclude that the observed moderation effects are caused by short-run contemporaneous impacts rather than long-run lagged impacts of the childhood variables. Although attempts have beenmade in this paper to try and separate the short-run from the long-run effects in several cases (e.g. by using only those aged21 and over in the analysis, by interacting the respondent’s unemployment with contemporaneous variables such as education,and by using a longer lagged childhood variables measured at aged 11), we acknowledge that the explanation of short-run con-temporaneous effects cannot be completely ruled out from the interpretation of our regression estimates. Hence, extra caremust be taken when interpreting these coefficients, and that future research should reinvestigate the issue using differentcohort data sets, such as the National Child Development Study (NCDS58) and British Cohort Study (BCS70), in order to tryand distinguish lagged effects of childhood characteristics from short-run contemporaneous effects more effectively.

More generally, the results presented here have shown how people’s psychological responses to economic shocks mightbe correlated with certain childhood variables. Future research should return to investigate what these other childhood char-acteristics are, how causal are their effects, and which of them are targetable by policy makers for interventions.

Acknowledgements

I have benefited from discussions with Andrew Clark, Francesca Cornaglia, Jan-Emmanuel DeNeve, Andrew Oswald, Rich-ard Layard, James Vernoit, Nele Warrinnier, and participants at the Well-being over Life-course workshop at the LSE and inMelbourne, Australia. I also thank the Department of Work and Pension and the U.S. National Institute of Aging (Grant NoR01AG040640) for providing the funding of this Project.

Appendix A

See Tables A1–A4.

Table A1The correlation matrix of the standardized average childhood variables (aged 11–15).

(i) (ii) (iii) (iv) (v) (vi) (vii) (viii) (ix) (x) (xi)

(i) 1.0000(ii) 0.1187* 1.0000(iii) �0.0388* 0.1436* 1.0000(iv) �0.0498* 0.1870* 0.5538* 1.0000(v) 0.0695* �0.0524* �0.0809* �0.0932* 1.0000(vi) 0.0171* �0.1541* �0.0494* �0.1154* 0.5439* 1.0000(vii) 0.0092 0.0836* 0.0715* 0.0375* �0.0261* �0.0156 1.0000(viii) �0.0173* 0.1015* 0.0563* 0.0981* �0.0548* �0.0260* 0.2196* 1.0000(ix) �0.0010 0.0308* 0.0052 0.0330* �0.0051 �0.0465* 0.0546* 0.0939* 1.0000(x) �0.0188* 0.0310* �0.0092 0.0011 �0.0236* �0.0075 0.0093 0.1209* 0.0952* 1.0000(xi) 0.0063 �0.0379* 0.0113 �0.0376* 0.1109* 0.0820* �0.1051* �0.1354* �0.3099* �0.2036* 1.0000

Note: (i) number of child’s close friends; (ii) frequency of child fighting with someone at school; (iii) frequency of child having arguments with father; (iv)frequency of having arguments with mother; (v) frequency of child talking to dad about things that matter; (vi) frequency of child talking to mum aboutthings that matter; (vii) father’s mental distress level; (viii) mother’s mental distress level; (ix) father’s unemployment; (x) mother’s unemployment; and(xi) log of real household income per annum.

* Indicates statistically significant at the 5% level.

Table A2FE Estimates of the control variables in Columns 3 & 7 of Table 2.

Mental well-being Life satisfaction

Age �0.339* �0.102[0.192] [0.078]

Age-squared 0.007* 0.003*

[0.004] [0.001]Self-employed 0.065 �0.116

[0.266] [0.126]Full-time student 0.045 0.028

[0.095] [0.042]Disabled �1.464** �0.397**

[0.646] [0.196]Other non-labour activities �0.099 0.025

[0.169] [0.075]Married 0.102 0.317***

[0.294] [0.097]Living as a couple 0.099 0.113**

[0.154] [0.056]Widowed/widower 0.515 �0.276

[1.251] [0.245]Divorced 1.215*** 0.711***

[0.329] [0.224]Separated �0.182 0.458*

[0.584] [0.243]Log of real household income per annum �0.027 �0.019

[0.045] [0.018]Education: A-level 0.368*** 0.102

[0.125] [0.075]Education: University degree 0.178 0.089

[0.160] [0.086]Health: poor �0.083 0.281

[0.455] [0.182]Health: fair 1.000** 0.588***

[0.447] [0.178]Health: good 1.399*** 0.785***

[0.450] [0.179]Health: excellent 1.674*** 0.973***

[0.451] [0.181]Number of children �0.298** �0.074

[0.147] [0.064]


Table A2 (continued)

Mental well-being Life satisfaction

Homeowner outright 0.325* �0.012[0.171] [0.065]

Mortgage/loan 0.284** �0.002[0.136] [0.051]

Unemployed � extraversion �0.005 0.002[0.015] [0.009]

Unemployed � agreeableness 0.018 0.005[0.033] [0.018]

Unemployed � openness 0.006 0.012[0.023] [0.011]

Unemployed � neuroticism �0.034** �0.006[0.017] [0.014]

Unemployed � conscientiousness 0.017 �0.014[0.022] [0.014]

Observations 8658 7996Number if individuals 1917 1893

Note: Cluster-robust standard errors (clustered at the individual level) are in parentheses.* <10%.

** <5%.*** <1%.

Table A3Fixed effects mental well-being regression equations with interaction effects between childhood characteristics at aged 11 or 15 and unemployment.

Variables All Men Women

a = 11 yearsold

a = 15 yearsold

a = 11 yearsold

a = 15 yearsold

a = 11 yearsold

a = 15 yearsold

Unemployment �0.876*** �0.962*** �1.091*** �1.197** �0.910** �0.985***

[0.245] [0.254] [0.420] [0.472] [0.361] [0.350]

Childhood characteristics at age = a interacted with unemploymentNumber of close friends 0.082 0.047 0.370 0.109 �0.013 �0.099

[0.243] [0.138] [0.270] [0.118] [0.353] [0.245]Frequency of fighting with someone �0.151 �0.145 �0.248 �0.232 �0.158 �0.281

[0.191] [0.167] [0.209] [0.197] [0.539] [0.353]Frequency of having arguments with father �0.663** �0.115 0.013 �0.011 �1.277** �0.114

[0.306] [0.238] [0.380] [0.341] [0.519] [0.356]Frequency of having arguments with mother 0.295 �0.093 �0.299 �0.166 0.992 0.012

[0.337] [0.262] [0.344] [0.305] [0.616] [0.411]Frequency of talking to dad about things that

matter0.464 �0.084 0.122 �0.065 0.623 �0.191[0.339] [0.204] [0.342] [0.289] [0.597] [0.320]

Frequency of talking to mum about things thatmatter

�0.573 0.090 �0.117 0.018 �0.962 0.125[0.385] [0.219] [0.334] [0.283] [0.731] [0.366]

Father’s mental distress (GHQ-12) 0.125 �0.025 �0.102 �0.287 0.277 0.323[0.122] [0.161] [0.138] [0.221] [0.199] [0.221]

Mother’s mental distress (GHQ-12) �0.396** �0.213 �0.062 �0.017 �0.563** �0.439*

[0.197] [0.156] [0.226] [0.165] [0.283] [0.263]Father’s unemployment 0.160 �0.041 0.136* �0.059 0.191 �0.044

[0.099] [0.087] [0.081] [0.108] [0.252] [0.148]Mother’s unemployment 0.230* 0.332* 0.208 0.260** 0.129 0.444

[0.130] [0.174] [0.212] [0.109] [0.242] [0.362]Log of real household income per annum 0.490 0.604* 0.340 0.549 0.533 0.800

[0.412] [0.314] [0.400] [0.387] [0.804] [0.590]

Exogenous control variables Yes Yes Yes Yes Yes YesSocio-economic control variables Yes Yes Yes Yes Yes YesObservations 8650 8650 4232 4232 4418 4418Number of individuals 1916 1916 971 971 945 945

Note: Cluster-robust standard errors (clustered at the individual level) are in parentheses. Same set of control variables as in Table 2. The intertemporalcorrelations in the selected childhood characteristics between aged 11 and 15 variables in the full sample are �0.068 for ‘‘Number of close friends’’; �0.016for ‘‘Frequency of fighting with someone’’; �0.028 for ‘‘Frequency of having arguments with father’’; �0.033 for ‘‘Frequency of having arguments withmother’’; 0.143 for ‘‘Frequency of talking to dad about things that matter’’; 0.140 for ‘‘Frequency of talking to mum about things that matter’’; 0.159 for‘‘Father’s mental distress (GHQ-12)’’; 0.146 for ‘‘Mother’s mental distress (GHQ-12)’’; 0.104 for ‘‘Father’s unemployment’’; 0.009 for ‘‘Mother’s unem-ployment’’; and �0.059 for ‘‘Log of real household income per annum’’.

* <10%.** <5%.

*** <1%.


Table A4Fixed effects life satisfaction regression equations with interaction effects between childhood characteristics at aged 11 or 15 and unemployment.

Variables All Men Women

a = 11 yearsold

a = 15 yearsold

a = 11 yearsold

a = 15 yearsold

a = 11 yearsold

a = 15 yearsold

Unemployment �0.265** �0.284** �0.134 �0.127 �0.298* �0.408**

[0.109] [0.116] [0.147] [0.170] [0.153] [0.164]

Childhood characteristics at age = a interacted with unemploymentNumber of close friends �0.022 0.063 �0.175 0.144** 0.147 �0.025

[0.104] [0.074] [0.137] [0.071] [0.150] [0.067]Frequency of fighting with someone �0.015 �0.014 �0.113 �0.102 �0.024 0.073

[0.102] [0.067] [0.116] [0.081] [0.270] [0.136]Frequency of having arguments with father �0.217** �0.029 �0.263* �0.087 �0.227 0.159

[0.103] [0.090] [0.155] [0.122] [0.151] [0.157]Frequency of having arguments with mother 0.034 �0.049 �0.122 �0.114 0.197 �0.052

[0.150] [0.105] [0.196] [0.123] [0.226] [0.166]Frequency of talking to dad about things that

matter0.408*** �0.035 0.480** 0.108 0.411* �0.093[0.142] [0.099] [0.191] [0.119] [0.220] [0.158]

Frequency of talking to mum about things thatmatter

�0.324* 0.060 �0.274 �0.080 �0.367 0.242*

[0.192] [0.094] [0.266] [0.130] [0.297] [0.139]Father’s mental distress (GHQ-12) 0.022 0.043 �0.047 �0.064 0.105 0.126

[0.064] [0.061] [0.081] [0.077] [0.099] [0.089]Mother’s mental distress (GHQ-12) �0.056 �0.102 0.019 �0.077 �0.067 �0.108

[0.082] [0.062] [0.112] [0.072] [0.111] [0.099]Father’s unemployment �0.046 0.034 �0.017 0.032 �0.103 0.053

[0.055] [0.048] [0.068] [0.060] [0.102] [0.089]Mother’s unemployment 0.082 0.006 �0.044 �0.083 0.181*** 0.132**

[0.055] [0.048] [0.047] [0.061] [0.069] [0.055]Log of real household income per annum 0.159 0.056 0.524** 0.097 �0.160 �0.172

[0.184] [0.130] [0.231] [0.143] [0.295] [0.248]

Exogenous control variables Yes Yes Yes Yes Yes YesSocio-economic control variables Yes Yes Yes Yes Yes YesObservations 7990 7990 3900 3900 4090 4090Number of individuals 1892 1892 956 956 936 936

Note: Cluster-robust standard errors (clustered at the individual level) are in parentheses. Same set of control variables as in Table 2. See Table A3’s note forthe intertemporal correlations between in the selected childhood characteristics between aged 11 and 15 variables in the full sample.

* <10%.** <5%.

*** <1%.


References

Becker, G. S. (1981). A treatise on the family. Cambridge, MA: Harvard University Press.Bonanno, G. A. (2004). Loss, trauma, and human resilience: Have we underestimated the human capacity to thrive after extremely aversive events?

American Psychologist, 59, 20–28.Bonanno, G. A. (2005). Resilience in the face of potential trauma. Current Directions in Psychological Science, 14, 135–138.Bonanno, G. A., Galea, S., Buccoarelli, A., & Vlahov, D. (2006). Psychological resilience after disaster: New York City in the aftermath of the September 11th

terrorist attack. Psychological resilience, 17, 181–186.Bonanno, G. A., Galea, S., Buccoarelli, A., & Vlahov, D. (2007). What predicts psychological resilience after disaster? The role of demographics, resources, and

life stress. Journal of Consulting and Clinical Psychology, 75(5), 671–682.Bonanno, G. A., Moskowitz, J. T., Papa, A., & Folkman, S. (2005). Resilience to loss in bereaved spouses, bereaved parents, and bereaved gay men. Journal of

Personality and Social Psychology, 88, 827–843.Bonanno, G. A., Rennicke, C., & Dekel, S. (2005). Self-enhancement among high-exposure survivors of the September 11th terrorist attack: Resilience or

social maladjustment? Journal of Personality and Social Psychology, 88, 984–998.Bonanno, G. A., Wortman, C. B., Lehman, D. R., Tweed, R. G., Haring, M., Sonnega, J., et al (2002). Resilience to loss and chronic grief: A prospective study from

preloss to 18-months postloss. Journal of Personality and Social Psychology, 83, 1150–1164.Boscarino, J. A., Galea, S., Adams, R. E., Ahern, J., Resnick, H., & Vlahov, D. (2004). Mental health service and medication use in New York City after the

September 11, 2001 terrorist attack. Psychiatric Services, 55, 274–283.Boyce, C. J., & Wood, A. M. (2011). Personality prior to disability determines adaptation: Agreeable individuals recover lost life satisfaction faster and more

completely. Psychological Science, 22, 183–191.Brennan, P. A., Brocque, R. L., & Hammen, C. (2003). Maternal depression, parent-child relationships, and resilient outcomes in adolescence. Journal of the

American Academy of Child and Adolescent Psychiatry, 42(12), 1469–1477.Cameron, C. A., & Miller, D. L. (2013). A practitioner’s guide to cluster-robust inference. Davis, mimeo: University of California.Carneiro, P. & Heckman, J. J. (2003). Human capital policy. IZA Discussion Papers No. 821. University of Bonn: Institute for the Study of Labor (IZA).Clark, A. E. (2003). Unemployment as a social norm: Psychological evidence from panel data. Journal of Labor Economics, 21(2), 323–351.Clark, A. E. & Lelkes, O. (2005). Deliver us from evil: religion as insurance. Paris School of Economics, manuscript.Clark, A. E., Diener, E., Georgellis, Y., & Lucas, R. E. (2008). Lags and leads in life satisfaction: A test of the baseline hypothesis. Economic Journal, 118(529),

F222–F243.Clark, A. E., Georgellis, Y., & Sanfey, P. J. (2001). Scarring: The psychological impact of past unemployment. Economica, 68, 221–241.Clark, A. E., Knabe, A., & Rätzel, S. (2009). Unemployment as a social norm in Germany. Schmollers Jahrbuch, 129(2), 251–260.Clark, A. E., & Oswald, A. J. (1994). Unhappiness and unemployment. Economic Journal, 104(424), 648–659.

http://refhub.elsevier.com/S0167-4870(14)00071-3/h0005




























Compas, B. E., Howell, D. C., Phares, V., Williams, R. A., & Guinta, C. T. (1989). Risk factors for emotional/behavioral problems in young adolescents: Aprospective analysis of adolescent and parental stress and symptoms. Journal of Consulting and Clinical Psychology, 57, 732–740.

Cunha, F., & Heckman, J. J. (2007). The technology of skill formation. No. w12840. National Bureau of Economic Research.Cunha, F., & Heckman, J. J. (2008). Formulating, identifying and estimating the technology of cognitive and non-cognitive skill formation. Journal of Human

Resources, 43(4), 738–782.Cunha, F., Heckman, J. J., & Schennach, S. M. (2010). Estimating the technology of cognitive and noncognitive skill formation. Econometrica, 78(3), 883–931.Darity, W. A., & Goldsmith, A. H. (1996). Social psychology, unemployment and macroeconomics. Journal of Economic Perspectives, 10(1), 121–140.Diener, E., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The satisfaction with life scale. Journal of Personality Assessment, 45, 1–5.DuMont, K. A., Widom, C. S., & Czaja, S. J. (2007). Predictors of resilience in abused and neglected children grown-up: The role of individual and

neighborhood characteristics. Child Abuse and Neglect, 31(3), 255–274.Eggers, A., Gaddy, C., & Graham, C. (2006). Well-being and unemployment in Russia in the 1990s: Can society’s suffering be individuals’ solace? Journal of

Socio-Economics, 35, 209–242.Enoch, M.-A. (2011). The role of early life stress as a predictor for alcohol and drug dependence. Psychopharmacology, 214, 17–31.Ferrer-i-Carbonell, A., & Frijters, P. (2004). How important is the methodology for the estimates of the determinants of happiness? Economic Journal,

114(497), 641–659.Flores, E., Cicchetti, D., & Rogosch, F. A. (2005). Predictors of resilience in maltreated and nonmaltreated Latino. Developmental Psychology, 41(2), 338–351.Flynn, R. J., & Biro, C. (1998). Comparing developmental outcomes for children in care with those for other children in Canada. Children and Society, 12,

228–233.Flynn, R. J., Ghazal, H., Legault, L., Vandermeulen, G., & Petrick, S. (2004). Use of population measures and norms to identify resilient outcomes in young

people in care: An exploratory study. Child and Family Social Work, 9, 65–79.Frederick, S. S., & Loewenstein, G. (1999). Hedonic adaptation. In D. Kahneman, E. Diener, & N. Schwarz (Eds.), The foundations of hedonic psychology

(pp. 302–329). New York, NY, US: Russell Sage Foundation.Freedy, J. R., Shaw, D. L., Jarrel, M. P., & Master, C. R. (1992). Toward an understanding of the psychological impact of disasters. Journal of Traumatic Stress, 5,

441–454.Frijters, P., Johnston, D. W. & Shields, M. A. (2014). Does childhood predict adult life satisfaction? Evidence from British Cohort Surveys. Economic Journal, in

press.Frijters, P., Johnston, D. W., & Shields, M. A. (2011). Life satisfaction dynamics with quarterly life event data. Scandinavian Journal of Economics, 113(1),

190–211.Galatzer-Levy, I., & Bonanno, G. A. (2012). Beyond normality in the study of bereavement: Heterogeneity in depression outcomes following loss in older

adults. Social Science and Medicine, 74(12), 1987–1994.Galea, S., Ahern, J., Resnick, H., Kilpatrick, D., Bucuvalas, M., Gold, J., et al (2002). Psychological sequelae of the September 11 terrorist attacks in New York

City. The New England Journal of Medicine, 346, 982–987.Graham, L., & Oswald, A. J. (2010). Hedonic capital, resilience and adaptation. Journal of Economic Behavior & Organization, 76, 372–384.Guthrie, E., Black, D., Bagalkote, H., Shaw, C., Campbell, M., & Creed, F. (1998). Psychological stress and burnout in medical students: A five-year prospective

longitudinal study. Journal of the Royal Society of Medicine, 91, 237–243.Heckman, J. J. (2000). Policies to foster human capital. Research in Economics, 54(1), 3–56.Hobfoll, S. E. (2002). Social and psychological resources and adaptation. Review of General Psychology, 6, 307–324.Kassenboehmer, S. C., & Haisken-DeNew, J. P. (2009). You’re fired! The causal negative effect of unemployment on life satisfaction. Economic Journal, 536(3),

448–462.Layard, R. (2006). Happiness and public policy: A challenge to the profession. Economic Journal, 116, C24–C33.Loewenstein, G., & Ubel, P. A. (2008). Hedonic adaptation and the role of decision and experience utility in public policy. Journal of Public Economics, 92(8–9),

1795–1810.Masten, A. S., Best, K. M., & Garmezy, N. (1990). Resilience and development: Contributions from the study of children who overcome adversity.

Development and Psychopathology, 2, 425–444.Oswald, A. J., & Powdthavee, N. (2008). Does happiness adapt? A longitudinal study of disability with implications for economists and judges. Journal of

Public Economics, 92(5–6), 1061–1077.Perez-Truglia, R. (2012). On the causes and consequences of hedonic adaptation. Journal of Economic Psychology, 33, 1182–1192.Powdthavee, N. (2007). Are there geographical variations in the psychological costs of unemployment in South Africa? Social Indicators Research, 80(3),

629–652.Powdthavee, N. (2012). Jobless, friendless and broke: What happens to different areas of life before and after unemployment? Economica, 79(315), 557–575.Powdthavee, N., & Vernoit, J. (2013). Parental unemployment and children’s happiness: A longitudinal study of young people’s well-being in unemployed

households. Labour Economics, 24, 253–263.Rayo, L., & Becker, G. S. (2007). Evolution efficiency and happiness. Journal of Political Economy, 115(2), 302–337.Taylor, M. F., Brice, J., Buck, N., & Prentice-Lane, E. (2002). British household panel survey user manual. Colchester: University of Essex.Todd, P. E., & Wolpin, K. I. (2007). The production of cognitive achievement in children: Home, school and racial test score gap. Journal of Human Capital, 1(1),

91–136.Winkelmann, L., & Winkelmann, R. (1998). Why are the unemployed so unhappy? Economica, 65(257), 1–15.























































what childhood characteristics predict psychological resilience to economic shocks in adulthood?

Documents